Extreme Value Analysis Model for Censored Sample and Its Application to the Measurement Data

Applicability of the extreme value analysis model calculating the parameter with the least square method (LSM) for a censored sample is investigated through a large scale Monte-Carlo simulation. The study shows that the LSM model with a jackknife method may yield under- or over-estimation for either the return value or its variance in the upper and double censoring cases with smaller sample size, whereas the model gives rise to precise estimates in the lower censoring and uncensored cases. Also, analyses for the measurement samples with different types of missing data reveal that the LSM model may produce an estimate of the return value comparable to the maximum likelihood method (MLM) model for a censored sample but a greater variance of the return value. This indicates that the LSM estimate is statistically less efficient than the MLM estimate.